Map fusion algorithm for multi-unmanned systems in GNSS-denied environment

The collaborative mapping of multiple-unmanned systems has been widely used in many fields such as national defense, digital mines, and smart cities. As a positioning device that unmanned systems rely heavily on in outdoor environments, intermittent denial of INS/GNSS will occur, which affects the positioning accuracy of unmanned systems and the accuracy of mapping. This paper proposes a multi-unmanned system map fusion algorithm based on lidar point cloud registration, which mainly includes point cloud preprocessing, point cloud coarse registration, fine registration and map fusion. At the initial moment of mapping, the noise points and outliers of the point clouds in the overlapping area are first removed by voxel filtering, ground segmentation and radius filtering; then 3D_HARRIS key points and corresponding FPFH feature descriptors are extracted for initial rough registration of sampling consistency ; Then take the coarse registration result as the initial transformation matrix, and use the voxel-based generalized iterative closest point method VGICP to perform fine registration to calculate the coordinate system transformation matrix between local maps; Fusion of local maps. The experimental results of map fusion in an outdoor GNSS-rejected environment show that the map fusion algorithm can construct a 3D point cloud map accurately and quickly.

I. Introduction

The use of multi-unmanned systems to collaboratively construct environmental maps can solve the problems of continuous accumulation of global errors in large-scale environmental mapping and excessive calculation, resulting in the failure of mapping. The use of multi-unmanned systems has greatly improved operations in extreme environments. Efficiency and safety have been widely used in open-pit mine unmanned driving [1][2], national defense and military [3], logistics and distribution and other fields. The map fusion problem involved in multiple unmanned systems aims to fuse and splicing the local submaps established by each unmanned system to form a complete global map. The core problem of map fusion is how to obtain the coordinate transformation matrix between the submaps.